A new wave of AI coverage tends to focus on which professions might eventually disappear. What a recently updated Stanford University study suggests is more specific, and in some ways more troubling: the disruption is not arriving evenly across age groups. It is concentrating at the starting line of careers, affecting workers who have not yet had the chance to build the kind of experience that seems to offer protection.
A Gap That Keeps Widening
Economists at Stanford University have been tracking employment trends across occupations rated by their exposure to AI disruption. Their August 2026 update to a paper titled “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence” finds that the pattern they identified earlier is not fading. It is deepening.
Workers aged 22 to 25 in the occupations most exposed to AI disruption are now employed at levels 19 percent below their peers in fields less affected by AI. A year earlier, that same gap stood at 13 percent. The trend is moving in one direction.
To measure this, the researchers drew on anonymized payroll data aggregated by ADP, a human resources management company, and rated occupational exposure to AI using two separate tools: an established labor market impact gauge and the Anthropic Economic Index, which tracks how different occupations actually use the Claude AI model in their day-to-day work. Google released a comparable report based on Gemini usage patterns around the same period.
Crucially, when the researchers looked at the economy as a whole, without separating by age, they found little to no difference in overall employment between the most and least AI-exposed occupations. The aggregate picture looks stable. The picture for young workers does not.
Automation Versus Augmentation: Not All AI Exposure Is the Same
Here is what most coverage of this study misses. The researchers did not treat AI exposure as a single uniform force. They drew on a distinction the Anthropic Economic Index makes between two types of AI use: “automative” queries, where AI fully replaces a task previously done by a human, and “augmentative” queries, where AI helps a human worker do their job more effectively.
Occupations like accountants and auditors, and receptionists and information clerks, ranked among those most susceptible to automation. Roles like chief executive and registered nurse ranked among those where AI is used primarily to augment human judgment and experience. The employment data follows this distinction closely. Jobs where automation is prevalent show the worst relative employment levels for young workers. Jobs where AI augments experienced workers show flat or rising employment.
The researchers also identified a related pattern around the type of knowledge a job requires. Entry-level roles tend to rely heavily on what they call “codified” knowledge: formal, standardized information that can be learned through education and written procedures. Senior roles tend to rely more on “tacit” knowledge, the kind built through years of practice, mentorship, and repeated exposure to real situations. AI, it appears, is more capable of handling codified knowledge than tacit knowledge. That makes entry-level workers in AI-exposed fields particularly vulnerable, not because they are less capable, but because the knowledge they bring to the table is the kind AI can most readily replicate.
One additional finding: the employment decline among young workers in AI-exposed fields is driven primarily by lower hiring rates, not by increased layoffs or resignations. Employers are simply bringing fewer new workers in.
The On-Ramp Problem
What makes this research significant is not just the numbers. It is the structural dynamic they point toward.
Lead researcher Erik Brynjolfsson, in an interview with The Washington Post, described the risk as a labor market that maintains its overall employment level while quietly closing the entry point for people starting their careers. The jobs held by workers already established in AI-exposed fields appear largely unaffected so far. The disruption is falling on those who have not yet entered.
This matters for how societies think about education, career pathways, and the long-term distribution of opportunity. If AI systematically reduces the availability of entry-level roles in certain fields, it does not just affect individual job seekers. It removes the training ground through which workers have historically developed the tacit knowledge that makes them valuable later in their careers. You cannot become a senior professional in a field if the junior roles that build toward that seniority are no longer available.
The researchers did find one partial buffer: occupations with a higher share of college graduates showed more muted differences between AI-exposed and less-exposed roles. In fields with fewer college graduates, the contrast was sharper, with the least AI-exposed occupations growing and the most exposed ones declining.
That finding does not resolve the problem. It adds a layer of complexity about who bears the cost.
In Short
A Stanford University study tracking employment data through 2026 finds that workers aged 22 to 25 in AI-exposed occupations are employed at levels 19 percent below peers in less-exposed fields, up from a 13 percent gap the previous year. The effect is driven by lower hiring, not layoffs, and is concentrated in roles that rely on codified, teachable knowledge rather than experience-based tacit knowledge. Occupations where AI automates tasks show worse outcomes than those where AI augments human workers. The broader economy looks stable in aggregate. The entry point to careers in certain fields does not.
Based on reporting from Ars Technica.